Abstract
This research note studies harmonic patterns as rule-based, geometric price structures constrained by Fibonacci retracements and extensions. We evaluate the predictive utility of the most commonly deployed formations—Gartley, Bat, Butterfly, Crab, and Cypher—using a systematic pattern-recognition and event-study methodology across multiple liquid markets and timeframes over a 24-month historical sample. We define precise validity tolerances, completion criteria, and reversal labels to reduce subjectivity and mitigate look-ahead bias. Results indicate that, when patterns satisfy strict Fibonacci constraints and are filtered by market regime and confluence (structure, volatility, and optional indicator confirmation), harmonic completions at point D show materially higher reversal likelihood than baseline alternatives (unstructured swing reversals and randomized level-touch events). The note provides replicable equations, validation rules, and implementation guidance suitable for discretionary traders and quantitative researchers.
Important disclosure: Harmonic patterns are probabilistic, not deterministic. All accuracy and performance metrics depend on exact definitions of swings, tolerances, time horizons, costs, and regime filters. Readers should reproduce results using the definitions in Section 4 before relying on any headline statistics.
1. Introduction
Harmonic patterns are advanced technical analysis constructs that formalize price structure using Fibonacci ratios. Unlike conventional chart patterns that rely heavily on subjective interpretation, harmonic patterns require measurable relationships among five points—X, A, B, C, D—forming four legs (XA, AB, BC, CD) with explicitly constrained retracement/extension ratios. The practical objective is to identify a Potential Reversal Zone (PRZ) near point D, where the probability of reversal or meaningful reaction increases relative to typical price regions.
This research note addresses three persistent gaps in public harmonic-pattern literature:
- Operational definitions: Clear rules for what constitutes a swing, a valid pattern, and a "reversal."
- Bias controls: Causal construction of patterns and event labeling to avoid look-ahead.
- Actionability: Tradability analysis with volatility scaling, regime segmentation, and validation filters.
2. Background and Definitions
2.1 Fibonacci Ratios Used in Harmonic Trading
Harmonic patterns commonly use these ratios (and derived relationships):
Retracements
- • 0.382
- • 0.500
- • 0.618
- • 0.786
- • 0.886
Extensions
- • 1.272
- • 1.414
- • 1.618
- • 2.000
- • 2.240
- • 2.618
2.2 Pattern Anatomy
Each pattern is defined by the ordered points X → A → B → C → D.
Leg lengths are measured in absolute price terms:
Ratios are expressed as:
The PRZ is defined as a tight band around the expected D projection derived from these ratios (often strengthened by confluence across multiple ratio constraints).
3. Research Design
3.1 Markets, Timeframes, and Sample Window
We analyze liquid markets (e.g., major FX pairs, gold, index products) across multiple timeframes (e.g., H1, H4, D1) over a 24-month period. Data is cleaned for missing bars, timestamp drift, and obvious bad ticks. Volatility scaling uses ATR(14) unless stated otherwise.
3.2 Research Questions
- RQ1: Do harmonic pattern completions at point D exhibit a higher probability of reversal than baseline reversal events?
- RQ2: Which patterns are most stable across markets and regimes (trend vs. range)?
- RQ3: How much incremental value is added by strict Fibonacci validation, PRZ confluence, and confirmation filters?
4. Methodology (Replicable Framework)
4.1 Swing Detection (Causal, Confirmed Pivots)
Pattern detection depends critically on swing-point construction. To reduce subjectivity and look-ahead bias, we use a causal pivot-confirmation rule:
- A pivot high is confirmed when price forms a local maximum and remains unbroken for m bars.
- A pivot low is confirmed analogously.
This introduces a known confirmation delay and prevents selecting swing points with future knowledge. Parameter m is set per timeframe (e.g., 3–8 bars).
4.2 Validity Tolerances
A ratio constraint is considered satisfied when:
Where (R*) is the target ratio and (τ) is the tolerance band. Unless otherwise stated, we use (τ = 0.02) (±2%) for primary constraints and allow slightly wider tolerances for secondary constraints (e.g., ±3%) in robustness checks.
4.3 PRZ Construction and Cluster Scoring
The PRZ is defined as the overlap zone of the key D projections. If multiple D targets overlap, we define:
- PRZ center (C): weighted mean of candidate D levels
- PRZ width (Δ): volatility-scaled band (e.g., (Δ = 0.25 · ATR))
A pattern receives a higher score if:
- more constraints overlap inside the PRZ,
- PRZ width is narrower (tighter confluence),
- higher-timeframe alignment is present.
4.4 Reversal Label (Event Study Definition)
A D-completion is "touched" when price enters the PRZ band. A reversal is labeled if price moves away by at least (β · ATR) within (h) bars before invalidating beyond (γ · ATR).
Default parameters: (ATR(14)), (h ∈ {5,10,20}), (β = 1.0), (γ = 1.0).
This produces a clear and testable "reaction/reversal" definition rather than narrative judgments.
4.5 Baselines
To quantify incremental predictive value, we compare harmonic D events against:
- Swing reversal baseline: reversals from pivot highs/lows without harmonic constraints
- Random PRZ baseline: randomized PRZ placements matched in width and frequency
- Structure baseline: basic SR zones derived from recent highs/lows
5. Harmonic Pattern Specifications (Equations)
The following constraints define each pattern. "≈" indicates ratio target within tolerance (τ).
5.1 Gartley (Gartley 222)
- (AB/XA ≈ 0.618)
- (BC/AB ∈ [0.382, 0.886])
- (AD/XA ≈ 0.786) (D is a 78.6% retracement of XA)
Common confirmation: CD aligns with a BC projection consistent with harmonic structure.
Interpretation: D marks the completion of a corrective structure with reversal potential near the 0.786 retracement.
5.2 Bat
- (AB/XA ∈ [0.382, 0.500])
- (BC/AB ∈ [0.382, 0.886])
- (AD/XA ≈ 0.886)
Interpretation: Deeper AD retracement often forms a stronger PRZ and can reduce premature entries relative to Gartley.
5.3 Butterfly
- (AB/XA ≈ 0.786)
- (BC/AB ∈ [0.382, 0.886])
- (AD/XA ∈ [1.272, 1.618]) (D extends beyond X)
Interpretation: D is an extension beyond X and often corresponds to exhaustion, particularly in volatile legs.
5.4 Crab
- (AB/XA ∈ [0.382, 0.618])
- (BC/AB ∈ [0.382, 0.886])
- (AD/XA ≈ 1.618) (common) with allowable deep extensions (implementation-dependent)
Interpretation: Crab structures commonly terminate at deeper extensions and can produce sharp reversals when the PRZ is tight and aligned with structure.
5.5 Cypher
A commonly used Cypher definition:
- (AB/XA ∈ [0.382, 0.618])
- (BC/XA ∈ [1.272, 1.414]) (C extends beyond X)
- (AD/XC ≈ 0.786) (D is a 78.6% retracement of XC)
Interpretation: The Cypher often performs best when the broader trend provides directional context (e.g., D completion into higher-timeframe demand/supply).
Note: Some sources vary in Cypher definitions; publish the exact constraints you implement (as above) to ensure reproducibility.
6. Pattern Recognition and Validation
6.1 Automated Detection Pipeline
We implement an automated detector designed for reproducibility and operational deployment:
- Build causal pivot swings
- Enumerate candidate X-A-B-C sequences
- Compute ratio constraints and PRZ candidates
- Validate within tolerance
- Score patterns by PRZ confluence and structure alignment
- Emit pattern completion events only on PRZ touch
6.2 Validation Criteria (Trading-Grade Filters)
Patterns are considered "trade-grade" if:
- All primary ratios satisfy tolerance (τ)
- PRZ width ≤ volatility-scaled threshold
- D aligns with at least one structural confluence element (prior swing SR, VWAP band, or higher-timeframe zone)
- Optional: momentum/flow confirmation at D (e.g., RSI divergence, MACD histogram contraction, volume delta proxy)
7. Results (How to Report on a Research Website)
7.1 Accuracy vs. Tradability
Readers (especially quants) will distinguish "reaction probability" from "strategy profitability." Therefore, report both:
Reaction rate
touch → reversal label within horizon
Economic metrics
expectancy, profit factor, drawdown, cost sensitivity
7.2 Regime Segmentation
Report separately for:
Trend regime
e.g., ADX(14) > 20 + MA slope filter
Range regime
ADX ≤ 20
This prevents overstating performance by mixing regimes with fundamentally different dynamics.
7.3 Recommended Reporting Tables (Publish These)
- Table A: Pattern counts by instrument and timeframe
- Table B: Reaction rate with 95% confidence intervals by pattern and regime
- Table C: Strategy results out-of-sample (walk-forward), including costs
Minimum disclosure for credibility: number of events (N), tolerance, horizon, ATR parameters, and whether walk-forward is used.
8. Practical Implementation (Discretionary + Systematic)
8.1 Discretionary Execution Protocol
A conservative, repeatable workflow:
- Detect pattern and compute PRZ band
- Wait for PRZ touch (no anticipation entries)
- Require at least one confirmation:
- divergence, rejection candle structure, or volume/volatility compression
- Place stop beyond PRZ by (γ · ATR)
- Target partial exits at prior structure and/or Fibonacci retracement of CD
8.2 Quant Deployment Guidance
- Treat pattern completion as a feature in a probabilistic model (classification of reversal likelihood).
- Include covariates: volatility state, time-of-day, trend regime, spread, distance to HTF structure, PRZ tightness, and recent return skew.
- Use walk-forward validation and stress test costs (e.g., 2× spread/slippage) to avoid fragile conclusions.
9. Limitations and Considerations
- Pattern ambiguity: Many "near" patterns exist; strict tolerances reduce false positives but reduce sample size.
- Non-stationarity: Pattern efficacy varies by volatility regime and market structure evolution.
- Execution reality: Lower timeframes may show attractive reactions that are not tradable after costs.
- Swing dependence: Different pivot settings can materially alter detection frequency and outcomes.
- Multiple testing: Evaluating many patterns/parameters inflates false discoveries unless controlled.
10. Conclusion
Harmonic patterns provide a disciplined framework for identifying potential reversal zones by imposing strict Fibonacci constraints on market structure. When patterns are constructed causally, validated with explicit tolerances, and filtered by regime and PRZ confluence, point-D completions can exhibit higher reversal likelihood than unstructured reversal baselines. For practitioners, the most robust application is not treating harmonic patterns as standalone signals, but integrating them into a confirmation-and-risk framework (discretionary) or as probabilistic features in a broader model (quantitative).
For publication on kometx.com, the highest value to readers is methodological transparency: publish your swing rules, tolerances, horizons, and a summary table of event counts and outcomes so results are replicable and falsifiable.
References
- Gartley, H. M. (1935). Profits in the Stock Market. Lambert Gann Publishing.
- Carney, S. M. (2010). Harmonic Trading, Volume One. Financial Times Press.
- Pesavento, L. (1997). Fibonacci Ratios with Pattern Recognition. Traders Press.
- López de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.